Shower reconstruction
All starts from one electron
Introduction
A lot of effort has been done, for OPERA experience and other studies with emulsion, to provide algorithms for shower reconstruction and electron/pion identification.
The starting algorithm is the one designed at Neuchatel by F. Juget:
Electron/pion separation with an Emulsion Cloud Chamber by using Neural Network - JINST 2 P01001 (2007)
A lot of variants have been designed since then, but the basic logic should still be valid. The algorithm follows these two steps:
Research for segments in a cone of given aperture angle (default 20 mrad), within also a cilynder of a given radius (default 800 micron). Basically there are both an angular and radial ranges.
The segments are accpeted according to acceptance in position DR and angle DT;
Application of Neural Network for electron/pion identification
Libraries
libShower/EdbShowerRec
This class is the most complete one. It is written to accept data in various formats, and then perform shower reconstruction on them. I usually pass the data by building EdbPVRec objects in them. I currently launch the following calls to this class:
Object instatiation: EdbShowerRec * showerrec = new EdbShowerRec();
Reset Arrays: showerrec->ResetInBTArray() and showerrec->ResetRecoShowerArray();
Load Initiator Basetracks: showerrec->SetInBTArray(myInBTArray);
This can be checked if worked with: showerrec->PrintInitiatorBTs();
Load PVRec container: showerrec->SetEdbPVRec(myEdbPVRec);
Set AliSub: showerrec->SetUseAliSub(0);
Start actual reconstruction: showerrec->Execute();
Check output of shower reconstruction: showerrec->PrintRecoShowerArray();
This operations are now done from the usual ScanSet interface with shower_reconstruction.C script. The parameters are set in a rootrc file, called showerrec.rootrc.
Default parameters:
showerrec.nbrick: 1
showerrec.cpcut: eN1==1&&eN2==1&&s1.eFlag>=0&&s2.eFlag>=0
showerrec.trkcus[0].Plate()<4 && s[0].Theta()<0.05
showerrec.ConeRadius: 1000
showerrec.ConeAngle: 0.04
showerrec.ConnectionDR: 150
showerrec.ConnectionDT: 0.15
showerrec.NPropagation: 3
showerrec.outdir: ..
showerrec.env: showerrec.rootrc
showerrec.EdbDebugLevel: 1
I have just discovered that NPropagation does not actually use the Plate Number, but the z position, assuming a reasonable dz:
if (dZ>(eAlgoParameterNPropagation*1273.0)+650.0) continue;
therefore, if a plate is missing, NPropagation needs to be increased, even if the plate is excluded by the EdbScanSet.
A tree called Shower.root should be produced. This contains the information about the segments contained in the shower and the ouput of the neural network.
The condition "eN1==1&&eN2==1&&s1.eFlag>=0&&s2.eFlag>=0" is necessary to select the best ranked base-tracks (only in case of data, not from simulation). Otherwise we have multiple base-tracks for the same micro-track, increasing unphysically our data sample with the combinations!
libShowRec/EdbShowRec
This class has been recently (2019) updated by Frank Meisel. It has a larger set of options with respect to the old one, allowing to compare between different parameters and compute efficiency from MC simulations.
From a practical point of view, to use it we need to:
Copy the folder $FEDRA_ROOT/src/libShowRec/ShowRec in our user directory
Modify it as we need for our data/simulation, then launch make to compile it
Launch it with ./ShowRec/ShowRec and all the options as in the help information.
Required input files:
A PVRec with the couples. It is needed to have set eEdbPVRec->eDescendingZ = 1, so the pattern with larger number has lower z;
A linked_tracks.root file, to take injectors from. -LT1 takes the first segment, -LT2 takes the last segment. It is not needed if -LT0 option is used, but this option requires a lot of time, since injectors are taken directly from the couples;
Additional input file:
ParameterSet tree: this allows to perform simulation with different sets of parameters (each entry is a different set). If the tree is not available, only one reconstruction with default values will be performed;
Particle Gun simulation text file: This text file contains the information from a MC simulation, to provide information about efficiency of reconstruction.
Example of working command (parameter set from entry 0 to entry 2 of parameter set tree):
It will provide as output many files: a shower tree file, an efficiency tree file, a histo file, and so on...
In order to the MC results to be reliable, you should provide the following variables:
MCEvt and MCTrack, as with the standard s.eMCEvt and s.MCTrack members;
PdgCode of the particle, as in s.Flag();
Momentum of the particle, stored in P();
Iniector track selection
Current shower reconstruction algorithms heavily depend on the choice of starter base-track (injiector of the shower). These candidate base-tracks are usually chosen as the first (or last) segment of the volume tracks in linked_tracks.root input file, according to our information.
For DESY19 testbeam, we have electron beam impinging on the start of the target. Therefore, we can ask the injector to be the first segment of the following subsample of tracks:
Note: sf[0].Theta() is the angle of the track projection, s[0].Theta() is the angle of the base-track. There is no significant difference on which one is used for this condition. However, DO NOT require sf[0].Plate()<4, since it has no effect at all (sf.Plate() always returns 0 for all segments!)
Last updated